Half-precision floating-point format

Last updated

In computing, half precision (sometimes called FP16 or float16) is a binary floating-point computer number format that occupies 16 bits (two bytes in modern computers) in computer memory. It is intended for storage of floating-point values in applications where higher precision is not essential, in particular image processing and neural networks.

Contents

Almost all modern uses follow the IEEE 754-2008 standard, where the 16-bit base-2 format is referred to as binary16, and the exponent uses 5 bits. This can express values in the range ±65,504, with the minimum value above 1 being 1 + 1/1024.

Depending on the computer, half-precision can be over an order of magnitude faster than double precision, e.g. 550 PFLOPS for half-precision vs 37 PFLOPS for double precision on one cloud provider. [1]

History

Several earlier 16-bit floating point formats have existed including that of Hitachi's HD61810 DSP of 1982 (a 4-bit exponent and a 12-bit mantissa), [2] Thomas J. Scott's WIF of 1991 (5 exponent bits, 10 mantissa bits) [3] and the 3dfx Voodoo Graphics processor of 1995 (same as Hitachi). [4]

ILM was searching for an image format that could handle a wide dynamic range, but without the hard drive and memory cost of single or double precision floating point. [5] The hardware-accelerated programmable shading group led by John Airey at SGI (Silicon Graphics) used the s10e5 data type in 1997 as part of the 'bali' design effort. This is described in a SIGGRAPH 2000 paper [6] (see section 4.3) and further documented in US patent 7518615. [7] It was popularized by its use in the open-source OpenEXR image format.

Nvidia and Microsoft defined the half datatype in the Cg language, released in early 2002, and implemented it in silicon in the GeForce FX, released in late 2002. [8] However, hardware support for accelerated 16-bit floating point was later dropped by Nvidia before being reintroduced in the Tegra X1 mobile GPU in 2015.

The F16C extension in 2012 allows x86 processors to convert half-precision floats to and from single-precision floats with a machine instruction.

IEEE 754 half-precision binary floating-point format: binary16

The IEEE 754 standard [9] specifies a binary16 as having the following format:

The format is laid out as follows:

IEEE 754r Half Floating Point Format.svg

The format is assumed to have an implicit lead bit with value 1 unless the exponent field is stored with all zeros. Thus, only 10 bits of the significand appear in the memory format but the total precision is 11 bits. In IEEE 754 parlance, there are 10 bits of significand, but there are 11 bits of significand precision (log10(211) ≈ 3.311 decimal digits, or 4 digits ± slightly less than 5 units in the last place).

Exponent encoding

The half-precision binary floating-point exponent is encoded using an offset-binary representation, with the zero offset being 15; also known as exponent bias in the IEEE 754 standard. [9]

Thus, as defined by the offset binary representation, in order to get the true exponent the offset of 15 has to be subtracted from the stored exponent.

The stored exponents 000002 and 111112 are interpreted specially.

ExponentSignificand = zeroSignificand ≠ zeroEquation
000002 zero, −0 subnormal numbers (−1)signbit × 2−14 × 0.significantbits2
000012, ..., 111102normalized value(−1)signbit × 2exponent−15 × 1.significantbits2
111112±infinity NaN (quiet, signalling)

The minimum strictly positive (subnormal) value is 2−24 ≈ 5.96 × 10−8. The minimum positive normal value is 2−14 ≈ 6.10 × 10−5. The maximum representable value is (2−2−10) × 215 = 65504.

Half precision examples

These examples are given in bit representation of the floating-point value. This includes the sign bit, (biased) exponent, and significand.

BinaryHexValueNotes
0 00000 000000000000000
0 00000 000000000100012−14 × (0 + 1/1024 ) ≈ 0.000000059604645smallest positive subnormal number
0 00000 111111111103ff2−14 × (0 + 1023/1024 ) ≈ 0.000060975552largest subnormal number
0 00001 000000000004002−14 × (1 + 0/1024 ) ≈ 0.00006103515625smallest positive normal number
0 01101 010101010135552−2 × (1 + 341/1024 ) ≈ 0.33325195nearest value to 1/3
0 01110 11111111113bff2−1 × (1 + 1023/1024 ) ≈ 0.99951172largest number less than one
0 01111 00000000003c0020 × (1 + 0/1024 ) = 1one
0 01111 00000000013c0120 × (1 + 1/1024 ) ≈ 1.00097656smallest number larger than one
0 11110 11111111117bff215 × (1 + 1023/1024 ) = 65504largest normal number
0 11111 00000000007c00∞infinity
1 00000 00000000008000−0
1 10000 0000000000c000−2
1 11111 0000000000fc00−∞negative infinity

By default, 1/3 rounds down like for double precision, because of the odd number of bits in the significand. The bits beyond the rounding point are 0101... which is less than 1/2 of a unit in the last place.

Precision limitations

MinMaxinterval
02−132−24
2−132−122−23
2−122−112−22
2−112−102−21
2−102−92−20
2−92−82−19
2−82−72−18
2−72−62−17
2−62−52−16
2−52−42−15
2−41/82−14
1/81/42−13
1/41/22−12
1/212−11
122−10
242−9
482−8
8162−7
16322−6
32642−5
641282−4
1282561/8
2565121/4
51210241/2
102420481
204840962
409681924
8192163848
163843276816
327686551932
65519∞∞

65519 is the largest number that will round to a finite number (65504), 65520 and larger will round to infinity. This is for round-to-even, other rounding strategies will change this cut-off.

ARM alternative half-precision

ARM processors support (via a floating point control register bit) an "alternative half-precision" format, which does away with the special case for an exponent value of 31 (111112). [10] It is almost identical to the IEEE format, but there is no encoding for infinity or NaNs; instead, an exponent of 31 encodes normalized numbers in the range 65536 to 131008.

Uses of half precision

Half precision is used in several computer graphics environments to store pixels, including MATLAB, OpenEXR, JPEG XR, GIMP, OpenGL, Vulkan, [11] Cg, Direct3D, and D3DX. The advantage over 8-bit or 16-bit integers is that the increased dynamic range allows for more detail to be preserved in highlights and shadows for images, and avoids gamma correction. The advantage over 32-bit single-precision floating point is that it requires half the storage and bandwidth (at the expense of precision and range). [5]

Half precision can be useful for mesh quantization. Mesh data is usually stored using 32-bit single precision floats for the vertices, however in some situations it is acceptable to reduce the precision to only 16-bit half precision, requiring only half the storage at the expense of some precision. Mesh quantization can also be done with 8-bit or 16-bit fixed precision depending on the requirements. [12]

Hardware and software for machine learning or neural networks tend to use half precision: such applications usually do a large amount of calculation, but don't require a high level of precision. Due to hardware typically not supporting 16-bit half precision floats, neural networks often use the bfloat16 format, which is the single precision float format truncated to 16 bits.

If the hardware has instructions to compute half-precision math, it is often faster than single or double precision. If the system has SIMD instructions that can handle multiple floating-point numbers within one instruction, half precision can be twice as fast by operating on twice as many numbers simultaneously. [13]

Support by programming languages

Zig provides support for half precisions with its f16 type. [14]

.NET 5 introduced half precision floating point numbers with the System.Half standard library type. [15] [16] As of January 2024, no .NET language (C#, F#, Visual Basic, and C++/CLI and C++/CX) has literals (e.g. in C#, 1.0f has type System.Single or 1.0m has type System.Decimal) or a keyword for the type. [17] [18] [19]

Hardware support

Several versions of the ARM architecture have support for half precision. [20]

Support for half precision in the x86 instruction set is specified in the F16C instruction set extension, first introduced in 2009 by AMD and fairly broadly adopted by AMD and Intel CPUs by 2012. This was further extended up the AVX-512_FP16 instruction set extension implemented in the Intel Sapphire Rapids processor. [21]

On RISC-V, the Zfh and Zfhmin extensions provide hardware support for 16-bit half precision floats. The Zfhmin extension is a minimal alternative to Zfh. [22]

On Power ISA, VSX and the not-yet-approved SVP64 extension provide hardware support for 16-bit half precision floats as of PowerISA v3.1B and later. [23] [24]

See also

Related Research Articles

<span class="mw-page-title-main">Floating-point arithmetic</span> Computer approximation for real numbers

In computing, floating-point arithmetic (FP) is arithmetic that represents subsets of real numbers using an integer with a fixed precision, called the significand, scaled by an integer exponent of a fixed base. Numbers of this form are called floating-point numbers. For example, 12.345 is a floating-point number in base ten with five digits of precision:

IEEE 754-1985 is a historic industry standard for representing floating-point numbers in computers, officially adopted in 1985 and superseded in 2008 by IEEE 754-2008, and then again in 2019 by minor revision IEEE 754-2019. During its 23 years, it was the most widely used format for floating-point computation. It was implemented in software, in the form of floating-point libraries, and in hardware, in the instructions of many CPUs and FPUs. The first integrated circuit to implement the draft of what was to become IEEE 754-1985 was the Intel 8087.

A computer number format is the internal representation of numeric values in digital device hardware and software, such as in programmable computers and calculators. Numerical values are stored as groupings of bits, such as bytes and words. The encoding between numerical values and bit patterns is chosen for convenience of the operation of the computer; the encoding used by the computer's instruction set generally requires conversion for external use, such as for printing and display. Different types of processors may have different internal representations of numerical values and different conventions are used for integer and real numbers. Most calculations are carried out with number formats that fit into a processor register, but some software systems allow representation of arbitrarily large numbers using multiple words of memory.

Double-precision floating-point format is a floating-point number format, usually occupying 64 bits in computer memory; it represents a wide dynamic range of numeric values by using a floating radix point.

In computer science, subnormal numbers are the subset of denormalized numbers that fill the underflow gap around zero in floating-point arithmetic. Any non-zero number with magnitude smaller than the smallest positive normal number is subnormal, while denormal can also refer to numbers outside that range.

The IEEE Standard for Floating-Point Arithmetic is a technical standard for floating-point arithmetic established in 1985 by the Institute of Electrical and Electronics Engineers (IEEE). The standard addressed many problems found in the diverse floating-point implementations that made them difficult to use reliably and portably. Many hardware floating-point units use the IEEE 754 standard.

The significand refers to the first (left) part of a number in scientific notation or related concepts in floating-point representation, consisting of its significant digits. Depending on the interpretation of the exponent, the significand may represent an integer or a fraction.

Hexadecimal floating point is a format for encoding floating-point numbers first introduced on the IBM System/360 computers, and supported on subsequent machines based on that architecture, as well as machines which were intended to be application-compatible with System/360.

In IEEE 754 floating-point numbers, the exponent is biased in the engineering sense of the word – the value stored is offset from the actual value by the exponent bias, also called a biased exponent. Biasing is done because exponents have to be signed values in order to be able to represent both tiny and huge values, but two's complement, the usual representation for signed values, would make comparison harder.

In C and related programming languages, long double refers to a floating-point data type that is often more precise than double precision though the language standard only requires it to be at least as precise as double. As with C's other floating-point types, it may not necessarily map to an IEEE format.

In computing, minifloats are floating-point values represented with very few bits. Predictably, they are not well suited for general-purpose numerical calculations. They are used for special purposes, most often in computer graphics, where iterations are small and precision has aesthetic effects. Machine learning also uses similar formats like bfloat16. Additionally, they are frequently encountered as a pedagogical tool in computer-science courses to demonstrate the properties and structures of floating-point arithmetic and IEEE 754 numbers.

Extended precision refers to floating-point number formats that provide greater precision than the basic floating-point formats. Extended precision formats support a basic format by minimizing roundoff and overflow errors in intermediate values of expressions on the base format. In contrast to extended precision, arbitrary-precision arithmetic refers to implementations of much larger numeric types using special software.

<span class="mw-page-title-main">CUDA</span> Parallel computing platform and programming model

CUDA is a proprietary and closed-source parallel computing platform and application programming interface (API) that allows software to use certain types of graphics processing units (GPUs) for general-purpose processing, an approach called general-purpose computing on GPUs (GPGPU). CUDA is a software layer that gives direct access to the GPU's virtual instruction set and parallel computational elements for the execution of compute kernels.

Decimal floating-point (DFP) arithmetic refers to both a representation and operations on decimal floating-point numbers. Working directly with decimal (base-10) fractions can avoid the rounding errors that otherwise typically occur when converting between decimal fractions and binary (base-2) fractions.

<span class="mw-page-title-main">OpenCL</span> Open standard for programming heterogenous computing systems, such as CPUs or GPUs

OpenCL is a framework for writing programs that execute across heterogeneous platforms consisting of central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs) and other processors or hardware accelerators. OpenCL specifies programming languages for programming these devices and application programming interfaces (APIs) to control the platform and execute programs on the compute devices. OpenCL provides a standard interface for parallel computing using task- and data-based parallelism.

In computing, quadruple precision is a binary floating-point–based computer number format that occupies 16 bytes with precision at least twice the 53-bit double precision.

Single-precision floating-point format is a computer number format, usually occupying 32 bits in computer memory; it represents a wide dynamic range of numeric values by using a floating radix point.

In computing, Microsoft Binary Format (MBF) is a format for floating-point numbers which was used in Microsoft's BASIC languages, including MBASIC, GW-BASIC and QuickBASIC prior to version 4.00.

In computing, octuple precision is a binary floating-point-based computer number format that occupies 32 bytes in computer memory. This 256-bit octuple precision is for applications requiring results in higher than quadruple precision. This format is rarely used and very few environments support it.

The bfloat16 floating-point format is a computer number format occupying 16 bits in computer memory; it represents a wide dynamic range of numeric values by using a floating radix point. This format is a shortened (16-bit) version of the 32-bit IEEE 754 single-precision floating-point format (binary32) with the intent of accelerating machine learning and near-sensor computing. It preserves the approximate dynamic range of 32-bit floating-point numbers by retaining 8 exponent bits, but supports only an 8-bit precision rather than the 24-bit significand of the binary32 format. More so than single-precision 32-bit floating-point numbers, bfloat16 numbers are unsuitable for integer calculations, but this is not their intended use. Bfloat16 is used to reduce the storage requirements and increase the calculation speed of machine learning algorithms.

References

  1. ↑ "About ABCI - About ABCI | ABCI". abci.ai. Retrieved 2019-10-06.
  2. ↑ "hitachi :: dataBooks :: HD61810 Digital Signal Processor Users Manual". Archive.org. Retrieved 2017-07-14.
  3. ↑ Scott, Thomas J. (March 1991). "Mathematics and computer science at odds over real numbers". Proceedings of the twenty-second SIGCSE technical symposium on Computer science education - SIGCSE '91. Vol. 23. pp. 130–139. doi: 10.1145/107004.107029 . ISBN   0897913779. S2CID   16648394.
  4. ↑ "/home/usr/bk/glide/docs2.3.1/GLIDEPGM.DOC". Gamers.org. Retrieved 2017-07-14.
  5. 1 2 "OpenEXR". OpenEXR. Retrieved 2017-07-14.
  6. ↑ Mark S. Peercy; Marc Olano; John Airey; P. Jeffrey Ungar. "Interactive Multi-Pass Programmable Shading" (PDF). People.csail.mit.edu. Retrieved 2017-07-14.
  7. ↑ "Patent US7518615 - Display system having floating point rasterization and floating point ... - Google Patents". Google.com. Retrieved 2017-07-14.
  8. ↑ "vs_2_sw". Cg 3.1 Toolkit Documentation. Nvidia. Retrieved 17 August 2016.
  9. 1 2 IEEE Standard for Floating-Point Arithmetic. IEEE STD 754-2019 (Revision of IEEE 754-2008). July 2019. pp. 1–84. doi:10.1109/ieeestd.2019.8766229. ISBN   978-1-5044-5924-2.
  10. ↑ "Half-precision floating-point number support". RealView Compilation Tools Compiler User Guide. 10 December 2010. Retrieved 2015-05-05.
  11. ↑ Garrard, Andrew. "10.1. 16-bit floating-point numbers". Khronos Data Format Specification v1.2 rev 1. Khronos. Retrieved 2023-08-05.
  12. ↑ "KHR_mesh_quantization". GitHub. Khronos Group. Retrieved 2023-07-02.
  13. ↑ Ho, Nhut-Minh; Wong, Weng-Fai (September 1, 2017). "Exploiting half precision arithmetic in Nvidia GPUs" (PDF). Department of Computer Science, National University of Singapore. Retrieved July 13, 2020. Nvidia recently introduced native half precision floating point support (FP16) into their Pascal GPUs. This was mainly motivated by the possibility that this will speed up data intensive and error tolerant applications in GPUs.
  14. ↑ "Floats". ziglang.org. Retrieved 7 January 2024.
  15. ↑ "Half Struct (System)". learn.microsoft.com. Retrieved 2024-02-01.
  16. ↑ Govindarajan, Prashanth (2020-08-31). "Introducing the Half type!". .NET Blog. Retrieved 2024-02-01.
  17. ↑ "Floating-point numeric types ― C# reference". learn.microsoft.com. 2022-09-29. Retrieved 2024-02-01.
  18. ↑ "Literals ― F# language reference". learn.microsoft.com. 2022-06-15. Retrieved 2024-02-01.
  19. ↑ "Data Type Summary — Visual Basic language reference". learn.microsoft.com. 2021-09-15. Retrieved 2024-02-01.
  20. ↑ "Half-precision floating-point number format". ARM Compiler armclang Reference Guide Version 6.7. ARM Developer. Retrieved 13 May 2022.
  21. ↑ Towner, Daniel. "Intel® Advanced Vector Extensions 512 - FP16 Instruction Set for Intel® Xeon® Processor Based Products" (PDF). Intel® Builders Programs. Retrieved 13 May 2022.
  22. ↑ "RISC-V Instruction Set Manual, Volume I: RISC-V User-Level ISA". Five EmbedDev. Retrieved 2023-07-02.
  23. ↑ "OPF_PowerISA_v3.1B.pdf". OpenPOWER Files. OpenPOWER Foundation. Retrieved 2023-07-02.
  24. ↑ "ls005.xlen.mdwn". libre-soc.org Git. Retrieved 2023-07-02.

Further reading

This page is based on this Wikipedia article
Text is available under the CC BY-SA 4.0 license; additional terms may apply.
Images, videos and audio are available under their respective licenses.